Constrained Score+(Local)Search Methods for Learning Bayesian Networks
The dominant approach for learning Bayesian networks from data is based on the use of a scoring metric, that evaluates the fitness of any given candidate network to the data, and a search procedure, that explores the space of possible solutions. The most used method inside this family is (iterated) hill climbing, because its good trade-off between CPU requirements, accuracy of the obtained model, and ease of implemetation. In this paper we focus on the searh space of dags and in the use of hill climbing as search engine. Our proposal consists in the reduction of the candidate dags or neighbors to be explored at each iteration, making the method more efficient on CPU time, but without decreasing the quality of the model discovered. Thus, initially the parent set for each variable is not restricted and so all the neighbors are explored, but during this exploration we take advantage of locally consistent metrics properties and remove some nodes from the set of candidate parents, constraining in this way the process for subsequent iterations. We show the benefits of our proposal by carrying out several experiments in three different domains.
KeywordsLocal Search Bayesian Network Hill Climbing Local Search Method Directed Cycle
Unable to display preview. Download preview PDF.
- 3.Beinlich, I.A., Suermondt, H.J., Chavez, R.M., Cooper, G.F.: The ALARM monitoring system: A case study with two probabilistic inference techniques for belief networks. In: Proc. of the 2nd European Conf. on Artificial Intelligence in Medicine, pp. 247–256 (1989)Google Scholar
- 5.Buntine, W.: Theory refinement of Bayesian networks. In: Proc. of the 7th Conf. on Uncertainty in Artificial Intelligence, pp. 52–60 (1991)Google Scholar
- 7.de Campos, L.M., Puerta, J.M.: Stochastic local and distributed search algorithms for learning belief networks. In: Proc. of 3rd Int. Symp. on Adaptive Systems: Evolutionary Computation and Probabilistic Graphical Model, pp. 109–115 (2001)Google Scholar
- 10.Chickering, D.M., Geiger, D., Heckerman, D.: Learning Bayesian networks is NP-Complete. In: Fisher, D., Lenz, H. (eds.) Learning from Data: Artificial Intelligence and Statistics V, pp. 121–130. Springer, Heidelberg (1996)Google Scholar
- 11.Friedman, N., Nachman, I., Peér, D.: Learning Bayesian networks from massive datasets: The ”sparse candidate” algorithm. In: Proc. of the 15th Conf. on Uncertainty in Artificial Intelligence, pp. 201–210 (1999)Google Scholar
- 14.Neapolitan, R.: Learning Bayesian Networks. Prentice-Hall, Englewood Cliffs (2003)Google Scholar
- 15.Pearl, J.: Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference. Morgan Kaufmann, San Mateo (1988)Google Scholar
- 17.Verma, T., Pearl, J.: Equivalence and synthesis of causal models. In: Proc. of the 6th Conf. on Uncertainty in Artificial Intelligence, pp. 220–227 (1991)Google Scholar